arch/x86/xen/enlighten.c

   1 /*
   2  * Core of Xen paravirt_ops implementation.
   3  *
   4  * This file contains the xen_paravirt_ops structure itself, and the
   5  * implementations for:
   6  * - privileged instructions
   7  * - interrupt flags
   8  * - segment operations
   9  * - booting and setup
  10  *
  11  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
  12  */
  13
  14 #include <linux/cpu.h>
  15 #include <linux/kernel.h>
  16 #include <linux/init.h>
  17 #include <linux/smp.h>
  18 #include <linux/preempt.h>
  19 #include <linux/hardirq.h>
  20 #include <linux/percpu.h>
  21 #include <linux/delay.h>
  22 #include <linux/start_kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/kprobes.h>
  25 #include <linux/bootmem.h>
  26 #include <linux/module.h>
  27 #include <linux/mm.h>
  28 #include <linux/page-flags.h>
  29 #include <linux/highmem.h>
  30 #include <linux/console.h>
  31 #include <linux/pci.h>
  32 #include <linux/gfp.h>
  33 #include <linux/memblock.h>
  34
  35 #include <xen/xen.h>
  36 #include <xen/interface/xen.h>
  37 #include <xen/interface/version.h>
  38 #include <xen/interface/physdev.h>
  39 #include <xen/interface/vcpu.h>
  40 #include <xen/interface/memory.h>
  41 #include <xen/features.h>
  42 #include <xen/page.h>
  43 #include <xen/hvm.h>
  44 #include <xen/hvc-console.h>
  45 #include <xen/acpi.h>
  46
  47 #include <asm/paravirt.h>
  48 #include <asm/apic.h>
  49 #include <asm/page.h>
  50 #include <asm/xen/pci.h>
  51 #include <asm/xen/hypercall.h>
  52 #include <asm/xen/hypervisor.h>
  53 #include <asm/fixmap.h>
  54 #include <asm/processor.h>
  55 #include <asm/proto.h>
  56 #include <asm/msr-index.h>
  57 #include <asm/traps.h>
  58 #include <asm/setup.h>
  59 #include <asm/desc.h>
  60 #include <asm/pgalloc.h>
  61 #include <asm/pgtable.h>
  62 #include <asm/tlbflush.h>
  63 #include <asm/reboot.h>
  64 #include <asm/stackprotector.h>
  65 #include <asm/hypervisor.h>
  66 #include <asm/mwait.h>
  67
  68 #ifdef CONFIG_ACPI
  69 #include <linux/acpi.h>
  70 #include <asm/acpi.h>
  71 #include <acpi/pdc_intel.h>
  72 #include <acpi/processor.h>
  73 #include <xen/interface/platform.h>
  74 #endif
  75
  76 #include "xen-ops.h"
  77 #include "mmu.h"
  78 #include "smp.h"
  79 #include "multicalls.h"
  80
  81 EXPORT_SYMBOL_GPL(hypercall_page);
  82
  83 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
  84 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
  85
  86 enum xen_domain_type xen_domain_type = XEN_NATIVE;
  87 EXPORT_SYMBOL_GPL(xen_domain_type);
  88
  89 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
  90 EXPORT_SYMBOL(machine_to_phys_mapping);
  91 unsigned long  machine_to_phys_nr;
  92 EXPORT_SYMBOL(machine_to_phys_nr);
  93
  94 struct start_info *xen_start_info;
  95 EXPORT_SYMBOL_GPL(xen_start_info);
  96
  97 struct shared_info xen_dummy_shared_info;
  98
  99 void *xen_initial_gdt;
 100
 101 RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
 102 __read_mostly int xen_have_vector_callback;
 103 EXPORT_SYMBOL_GPL(xen_have_vector_callback);
 104
 105 /*
 106  * Point at some empty memory to start with. We map the real shared_info
 107  * page as soon as fixmap is up and running.
 108  */
 109 struct shared_info *HYPERVISOR_shared_info = (void *)&xen_dummy_shared_info;
 110
 111 /*
 112  * Flag to determine whether vcpu info placement is available on all
 113  * VCPUs.  We assume it is to start with, and then set it to zero on
 114  * the first failure.  This is because it can succeed on some VCPUs
 115  * and not others, since it can involve hypervisor memory allocation,
 116  * or because the guest failed to guarantee all the appropriate
 117  * constraints on all VCPUs (ie buffer can't cross a page boundary).
 118  *
 119  * Note that any particular CPU may be using a placed vcpu structure,
 120  * but we can only optimise if the all are.
 121  *
 122  * 0: not available, 1: available
 123  */
 124 static int have_vcpu_info_placement = 1;
 125
 126 static void clamp_max_cpus(void)
 127 {
 128 #ifdef CONFIG_SMP
 129         if (setup_max_cpus > MAX_VIRT_CPUS)
 130                 setup_max_cpus = MAX_VIRT_CPUS;
 131 #endif
 132 }
 133
 134 static void xen_vcpu_setup(int cpu)
 135 {
 136         struct vcpu_register_vcpu_info info;
 137         int err;
 138         struct vcpu_info *vcpup;
 139
 140         BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
 141
 142         if (cpu < MAX_VIRT_CPUS)
 143                 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
 144
 145         if (!have_vcpu_info_placement) {
 146                 if (cpu >= MAX_VIRT_CPUS)
 147                         clamp_max_cpus();
 148                 return;
 149         }
 150
 151         vcpup = &per_cpu(xen_vcpu_info, cpu);
 152         info.mfn = arbitrary_virt_to_mfn(vcpup);
 153         info.offset = offset_in_page(vcpup);
 154
 155         /* Check to see if the hypervisor will put the vcpu_info
 156            structure where we want it, which allows direct access via
 157            a percpu-variable. */
 158         err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
 159
 160         if (err) {
 161                 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
 162                 have_vcpu_info_placement = 0;
 163                 clamp_max_cpus();
 164         } else {
 165                 /* This cpu is using the registered vcpu info, even if
 166                    later ones fail to. */
 167                 per_cpu(xen_vcpu, cpu) = vcpup;
 168         }
 169 }
 170
 171 /*
 172  * On restore, set the vcpu placement up again.
 173  * If it fails, then we're in a bad state, since
 174  * we can't back out from using it...
 175  */
 176 void xen_vcpu_restore(void)
 177 {
 178         int cpu;
 179
 180         for_each_online_cpu(cpu) {
 181                 bool other_cpu = (cpu != smp_processor_id());
 182
 183                 if (other_cpu &&
 184                     HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
 185                         BUG();
 186
 187                 xen_setup_runstate_info(cpu);
 188
 189                 if (have_vcpu_info_placement)
 190                         xen_vcpu_setup(cpu);
 191
 192                 if (other_cpu &&
 193                     HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
 194                         BUG();
 195         }
 196 }
 197
 198 static void __init xen_banner(void)
 199 {
 200         unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
 201         struct xen_extraversion extra;
 202         HYPERVISOR_xen_version(XENVER_extraversion, &extra);
 203
 204         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
 205                pv_info.name);
 206         printk(KERN_INFO "Xen version: %d.%d%s%s\n",
 207                version >> 16, version & 0xffff, extra.extraversion,
 208                xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
 209 }
 210
 211 static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
 212 static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
 213
 214 static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
 215 static __read_mostly unsigned int cpuid_leaf5_ecx_val;
 216 static __read_mostly unsigned int cpuid_leaf5_edx_val;
 217
 218 static void xen_cpuid(unsigned int *ax, unsigned int *bx,
 219                       unsigned int *cx, unsigned int *dx)
 220 {
 221         unsigned maskebx = ~0;
 222         unsigned maskecx = ~0;
 223         unsigned maskedx = ~0;
 224         unsigned setecx = 0;
 225         /*
 226          * Mask out inconvenient features, to try and disable as many
 227          * unsupported kernel subsystems as possible.
 228          */
 229         switch (*ax) {
 230         case 1:
 231                 maskecx = cpuid_leaf1_ecx_mask;
 232                 setecx = cpuid_leaf1_ecx_set_mask;
 233                 maskedx = cpuid_leaf1_edx_mask;
 234                 break;
 235
 236         case CPUID_MWAIT_LEAF:
 237                 /* Synthesize the values.. */
 238                 *ax = 0;
 239                 *bx = 0;
 240                 *cx = cpuid_leaf5_ecx_val;
 241                 *dx = cpuid_leaf5_edx_val;
 242                 return;
 243
 244         case 0xb:
 245                 /* Suppress extended topology stuff */
 246                 maskebx = 0;
 247                 break;
 248         }
 249
 250         asm(XEN_EMULATE_PREFIX "cpuid"
 251                 : "=a" (*ax),
 252                   "=b" (*bx),
 253                   "=c" (*cx),
 254                   "=d" (*dx)
 255                 : "0" (*ax), "2" (*cx));
 256
 257         *bx &= maskebx;
 258         *cx &= maskecx;
 259         *cx |= setecx;
 260         *dx &= maskedx;
 261
 262 }
 263
 264 static bool __init xen_check_mwait(void)
 265 {
 266 #ifdef CONFIG_ACPI
 267         struct xen_platform_op op = {
 268                 .cmd                    = XENPF_set_processor_pminfo,
 269                 .u.set_pminfo.id        = -1,
 270                 .u.set_pminfo.type      = XEN_PM_PDC,
 271         };
 272         uint32_t buf[3];
 273         unsigned int ax, bx, cx, dx;
 274         unsigned int mwait_mask;
 275
 276         /* We need to determine whether it is OK to expose the MWAIT
 277          * capability to the kernel to harvest deeper than C3 states from ACPI
 278          * _CST using the processor_harvest_xen.c module. For this to work, we
 279          * need to gather the MWAIT_LEAF values (which the cstate.c code
 280          * checks against). The hypervisor won't expose the MWAIT flag because
 281          * it would break backwards compatibility; so we will find out directly
 282          * from the hardware and hypercall.
 283          */
 284         if (!xen_initial_domain())
 285                 return false;
 286
 287         ax = 1;
 288         cx = 0;
 289
 290         native_cpuid(&ax, &bx, &cx, &dx);
 291
 292         mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
 293                      (1 << (X86_FEATURE_MWAIT % 32));
 294
 295         if ((cx & mwait_mask) != mwait_mask)
 296                 return false;
 297
 298         /* We need to emulate the MWAIT_LEAF and for that we need both
 299          * ecx and edx. The hypercall provides only partial information.
 300          */
 301
 302         ax = CPUID_MWAIT_LEAF;
 303         bx = 0;
 304         cx = 0;
 305         dx = 0;
 306
 307         native_cpuid(&ax, &bx, &cx, &dx);
 308
 309         /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
 310          * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
 311          */
 312         buf[0] = ACPI_PDC_REVISION_ID;
 313         buf[1] = 1;
 314         buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
 315
 316         set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
 317
 318         if ((HYPERVISOR_dom0_op(&op) == 0) &&
 319             (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
 320                 cpuid_leaf5_ecx_val = cx;
 321                 cpuid_leaf5_edx_val = dx;
 322         }
 323         return true;
 324 #else
 325         return false;
 326 #endif
 327 }
 328 static void __init xen_init_cpuid_mask(void)
 329 {
 330         unsigned int ax, bx, cx, dx;
 331         unsigned int xsave_mask;
 332
 333         cpuid_leaf1_edx_mask =
 334                 ~((1 << X86_FEATURE_MCE)  |  /* disable MCE */
 335                   (1 << X86_FEATURE_MCA)  |  /* disable MCA */
 336                   (1 << X86_FEATURE_MTRR) |  /* disable MTRR */
 337                   (1 << X86_FEATURE_ACC));   /* thermal monitoring */
 338
 339         if (!xen_initial_domain())
 340                 cpuid_leaf1_edx_mask &=
 341                         ~((1 << X86_FEATURE_APIC) |  /* disable local APIC */
 342                           (1 << X86_FEATURE_ACPI));  /* disable ACPI */
 343         ax = 1;
 344         cx = 0;
 345         xen_cpuid(&ax, &bx, &cx, &dx);
 346
 347         xsave_mask =
 348                 (1 << (X86_FEATURE_XSAVE % 32)) |
 349                 (1 << (X86_FEATURE_OSXSAVE % 32));
 350
 351         /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
 352         if ((cx & xsave_mask) != xsave_mask)
 353                 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
 354
 355         if (xen_check_mwait())
 356                 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
 357 }
 358
 359 static void xen_set_debugreg(int reg, unsigned long val)
 360 {
 361         HYPERVISOR_set_debugreg(reg, val);
 362 }
 363
 364 static unsigned long xen_get_debugreg(int reg)
 365 {
 366         return HYPERVISOR_get_debugreg(reg);
 367 }
 368
 369 static void xen_end_context_switch(struct task_struct *next)
 370 {
 371         xen_mc_flush();
 372         paravirt_end_context_switch(next);
 373 }
 374
 375 static unsigned long xen_store_tr(void)
 376 {
 377         return 0;
 378 }
 379
 380 /*
 381  * Set the page permissions for a particular virtual address.  If the
 382  * address is a vmalloc mapping (or other non-linear mapping), then
 383  * find the linear mapping of the page and also set its protections to
 384  * match.
 385  */
 386 static void set_aliased_prot(void *v, pgprot_t prot)
 387 {
 388         int level;
 389         pte_t *ptep;
 390         pte_t pte;
 391         unsigned long pfn;
 392         struct page *page;
 393
 394         ptep = lookup_address((unsigned long)v, &level);
 395         BUG_ON(ptep == NULL);
 396
 397         pfn = pte_pfn(*ptep);
 398         page = pfn_to_page(pfn);
 399
 400         pte = pfn_pte(pfn, prot);
 401
 402         if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
 403                 BUG();
 404
 405         if (!PageHighMem(page)) {
 406                 void *av = __va(PFN_PHYS(pfn));
 407
 408                 if (av != v)
 409                         if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
 410                                 BUG();
 411         } else
 412                 kmap_flush_unused();
 413 }
 414
 415 static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
 416 {
 417         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
 418         int i;
 419
 420         for(i = 0; i < entries; i += entries_per_page)
 421                 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
 422 }
 423
 424 static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
 425 {
 426         const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
 427         int i;
 428
 429         for(i = 0; i < entries; i += entries_per_page)
 430                 set_aliased_prot(ldt + i, PAGE_KERNEL);
 431 }
 432
 433 static void xen_set_ldt(const void *addr, unsigned entries)
 434 {
 435         struct mmuext_op *op;
 436         struct multicall_space mcs = xen_mc_entry(sizeof(*op));
 437
 438         trace_xen_cpu_set_ldt(addr, entries);
 439
 440         op = mcs.args;
 441         op->cmd = MMUEXT_SET_LDT;
 442         op->arg1.linear_addr = (unsigned long)addr;
 443         op->arg2.nr_ents = entries;
 444
 445         MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
 446
 447         xen_mc_issue(PARAVIRT_LAZY_CPU);
 448 }
 449
 450 static void xen_load_gdt(const struct desc_ptr *dtr)
 451 {
 452         unsigned long va = dtr->address;
 453         unsigned int size = dtr->size + 1;
 454         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 455         unsigned long frames[pages];
 456         int f;
 457
 458         /*
 459          * A GDT can be up to 64k in size, which corresponds to 8192
 460          * 8-byte entries, or 16 4k pages..
 461          */
 462
 463         BUG_ON(size > 65536);
 464         BUG_ON(va & ~PAGE_MASK);
 465
 466         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
 467                 int level;
 468                 pte_t *ptep;
 469                 unsigned long pfn, mfn;
 470                 void *virt;
 471
 472                 /*
 473                  * The GDT is per-cpu and is in the percpu data area.
 474                  * That can be virtually mapped, so we need to do a
 475                  * page-walk to get the underlying MFN for the
 476                  * hypercall.  The page can also be in the kernel's
 477                  * linear range, so we need to RO that mapping too.
 478                  */
 479                 ptep = lookup_address(va, &level);
 480                 BUG_ON(ptep == NULL);
 481
 482                 pfn = pte_pfn(*ptep);
 483                 mfn = pfn_to_mfn(pfn);
 484                 virt = __va(PFN_PHYS(pfn));
 485
 486                 frames[f] = mfn;
 487
 488                 make_lowmem_page_readonly((void *)va);
 489                 make_lowmem_page_readonly(virt);
 490         }
 491
 492         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
 493                 BUG();
 494 }
 495
 496 /*
 497  * load_gdt for early boot, when the gdt is only mapped once
 498  */
 499 static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
 500 {
 501         unsigned long va = dtr->address;
 502         unsigned int size = dtr->size + 1;
 503         unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
 504         unsigned long frames[pages];
 505         int f;
 506
 507         /*
 508          * A GDT can be up to 64k in size, which corresponds to 8192
 509          * 8-byte entries, or 16 4k pages..
 510          */
 511
 512         BUG_ON(size > 65536);
 513         BUG_ON(va & ~PAGE_MASK);
 514
 515         for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
 516                 pte_t pte;
 517                 unsigned long pfn, mfn;
 518
 519                 pfn = virt_to_pfn(va);
 520                 mfn = pfn_to_mfn(pfn);
 521
 522                 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
 523
 524                 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
 525                         BUG();
 526
 527                 frames[f] = mfn;
 528         }
 529
 530         if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
 531                 BUG();
 532 }
 533
 534 static void load_TLS_descriptor(struct thread_struct *t,
 535                                 unsigned int cpu, unsigned int i)
 536 {
 537         struct desc_struct *gdt = get_cpu_gdt_table(cpu);
 538         xmaddr_t maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
 539         struct multicall_space mc = __xen_mc_entry(0);
 540
 541         MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
 542 }
 543
 544 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
 545 {
 546         /*
 547          * XXX sleazy hack: If we're being called in a lazy-cpu zone
 548          * and lazy gs handling is enabled, it means we're in a
 549          * context switch, and %gs has just been saved.  This means we
 550          * can zero it out to prevent faults on exit from the
 551          * hypervisor if the next process has no %gs.  Either way, it
 552          * has been saved, and the new value will get loaded properly.
 553          * This will go away as soon as Xen has been modified to not
 554          * save/restore %gs for normal hypercalls.
 555          *
 556          * On x86_64, this hack is not used for %gs, because gs points
 557          * to KERNEL_GS_BASE (and uses it for PDA references), so we
 558          * must not zero %gs on x86_64
 559          *
 560          * For x86_64, we need to zero %fs, otherwise we may get an
 561          * exception between the new %fs descriptor being loaded and
 562          * %fs being effectively cleared at __switch_to().
 563          */
 564         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
 565 #ifdef CONFIG_X86_32
 566                 lazy_load_gs(0);
 567 #else
 568                 loadsegment(fs, 0);
 569 #endif
 570         }
 571
 572         xen_mc_batch();
 573
 574         load_TLS_descriptor(t, cpu, 0);
 575         load_TLS_descriptor(t, cpu, 1);
 576         load_TLS_descriptor(t, cpu, 2);
 577
 578         xen_mc_issue(PARAVIRT_LAZY_CPU);
 579 }
 580
 581 #ifdef CONFIG_X86_64
 582 static void xen_load_gs_index(unsigned int idx)
 583 {
 584         if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
 585                 BUG();
 586 }
 587 #endif
 588
 589 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
 590                                 const void *ptr)
 591 {
 592         xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
 593         u64 entry = *(u64 *)ptr;
 594
 595         trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
 596
 597         preempt_disable();
 598
 599         xen_mc_flush();
 600         if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
 601                 BUG();
 602
 603         preempt_enable();
 604 }
 605
 606 static int cvt_gate_to_trap(int vector, const gate_desc *val,
 607                             struct trap_info *info)
 608 {
 609         unsigned long addr;
 610
 611         if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
 612                 return 0;
 613
 614         info->vector = vector;
 615
 616         addr = gate_offset(*val);
 617 #ifdef CONFIG_X86_64
 618         /*
 619          * Look for known traps using IST, and substitute them
 620          * appropriately.  The debugger ones are the only ones we care
 621          * about.  Xen will handle faults like double_fault and
 622          * machine_check, so we should never see them.  Warn if
 623          * there's an unexpected IST-using fault handler.
 624          */
 625         if (addr == (unsigned long)debug)
 626                 addr = (unsigned long)xen_debug;
 627         else if (addr == (unsigned long)int3)
 628                 addr = (unsigned long)xen_int3;
 629         else if (addr == (unsigned long)stack_segment)
 630                 addr = (unsigned long)xen_stack_segment;
 631         else if (addr == (unsigned long)double_fault ||
 632                  addr == (unsigned long)nmi) {
 633                 /* Don't need to handle these */
 634                 return 0;
 635 #ifdef CONFIG_X86_MCE
 636         } else if (addr == (unsigned long)machine_check) {
 637                 return 0;
 638 #endif
 639         } else {
 640                 /* Some other trap using IST? */
 641                 if (WARN_ON(val->ist != 0))
 642                         return 0;
 643         }
 644 #endif  /* CONFIG_X86_64 */
 645         info->address = addr;
 646
 647         info->cs = gate_segment(*val);
 648         info->flags = val->dpl;
 649         /* interrupt gates clear IF */
 650         if (val->type == GATE_INTERRUPT)
 651                 info->flags |= 1 << 2;
 652
 653         return 1;
 654 }
 655
 656 /* Locations of each CPU's IDT */
 657 static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
 658
 659 /* Set an IDT entry.  If the entry is part of the current IDT, then
 660    also update Xen. */
 661 static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
 662 {
 663         unsigned long p = (unsigned long)&dt[entrynum];
 664         unsigned long start, end;
 665
 666         trace_xen_cpu_write_idt_entry(dt, entrynum, g);
 667
 668         preempt_disable();
 669
 670         start = __this_cpu_read(idt_desc.address);
 671         end = start + __this_cpu_read(idt_desc.size) + 1;
 672
 673         xen_mc_flush();
 674
 675         native_write_idt_entry(dt, entrynum, g);
 676
 677         if (p >= start && (p + 8) <= end) {
 678                 struct trap_info info[2];
 679
 680                 info[1].address = 0;
 681
 682                 if (cvt_gate_to_trap(entrynum, g, &info[0]))
 683                         if (HYPERVISOR_set_trap_table(info))
 684                                 BUG();
 685         }
 686
 687         preempt_enable();
 688 }
 689
 690 static void xen_convert_trap_info(const struct desc_ptr *desc,
 691                                   struct trap_info *traps)
 692 {
 693         unsigned in, out, count;
 694
 695         count = (desc->size+1) / sizeof(gate_desc);
 696         BUG_ON(count > 256);
 697
 698         for (in = out = 0; in < count; in++) {
 699                 gate_desc *entry = (gate_desc*)(desc->address) + in;
 700
 701                 if (cvt_gate_to_trap(in, entry, &traps[out]))
 702                         out++;
 703         }
 704         traps[out].address = 0;
 705 }
 706
 707 void xen_copy_trap_info(struct trap_info *traps)
 708 {
 709         const struct desc_ptr *desc = &__get_cpu_var(idt_desc);
 710
 711         xen_convert_trap_info(desc, traps);
 712 }
 713
 714 /* Load a new IDT into Xen.  In principle this can be per-CPU, so we
 715    hold a spinlock to protect the static traps[] array (static because
 716    it avoids allocation, and saves stack space). */
 717 static void xen_load_idt(const struct desc_ptr *desc)
 718 {
 719         static DEFINE_SPINLOCK(lock);
 720         static struct trap_info traps[257];
 721
 722         trace_xen_cpu_load_idt(desc);
 723
 724         spin_lock(&lock);
 725
 726         __get_cpu_var(idt_desc) = *desc;
 727
 728         xen_convert_trap_info(desc, traps);
 729
 730         xen_mc_flush();
 731         if (HYPERVISOR_set_trap_table(traps))
 732                 BUG();
 733
 734         spin_unlock(&lock);
 735 }
 736
 737 /* Write a GDT descriptor entry.  Ignore LDT descriptors, since
 738    they're handled differently. */
 739 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
 740                                 const void *desc, int type)
 741 {
 742         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
 743
 744         preempt_disable();
 745
 746         switch (type) {
 747         case DESC_LDT:
 748         case DESC_TSS:
 749                 /* ignore */
 750                 break;
 751
 752         default: {
 753                 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
 754
 755                 xen_mc_flush();
 756                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
 757                         BUG();
 758         }
 759
 760         }
 761
 762         preempt_enable();
 763 }
 764
 765 /*
 766  * Version of write_gdt_entry for use at early boot-time needed to
 767  * update an entry as simply as possible.
 768  */
 769 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
 770                                             const void *desc, int type)
 771 {
 772         trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
 773
 774         switch (type) {
 775         case DESC_LDT:
 776         case DESC_TSS:
 777                 /* ignore */
 778                 break;
 779
 780         default: {
 781                 xmaddr_t maddr = virt_to_machine(&dt[entry]);
 782
 783                 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
 784                         dt[entry] = *(struct desc_struct *)desc;
 785         }
 786
 787         }
 788 }
 789
 790 static void xen_load_sp0(struct tss_struct *tss,
 791                          struct thread_struct *thread)
 792 {
 793         struct multicall_space mcs;
 794
 795         mcs = xen_mc_entry(0);
 796         MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
 797         xen_mc_issue(PARAVIRT_LAZY_CPU);
 798 }
 799
 800 static void xen_set_iopl_mask(unsigned mask)
 801 {
 802         struct physdev_set_iopl set_iopl;
 803
 804         /* Force the change at ring 0. */
 805         set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
 806         HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
 807 }
 808
 809 static void xen_io_delay(void)
 810 {
 811 }
 812
 813 #ifdef CONFIG_X86_LOCAL_APIC
 814 static u32 xen_apic_read(u32 reg)
 815 {
 816         return 0;
 817 }
 818
 819 static void xen_apic_write(u32 reg, u32 val)
 820 {
 821         /* Warn to see if there's any stray references */
 822         WARN_ON(1);
 823 }
 824
 825 static u64 xen_apic_icr_read(void)
 826 {
 827         return 0;
 828 }
 829
 830 static void xen_apic_icr_write(u32 low, u32 id)
 831 {
 832         /* Warn to see if there's any stray references */
 833         WARN_ON(1);
 834 }
 835
 836 static void xen_apic_wait_icr_idle(void)
 837 {
 838         return;
 839 }
 840
 841 static u32 xen_safe_apic_wait_icr_idle(void)
 842 {
 843         return 0;
 844 }
 845
 846 static void set_xen_basic_apic_ops(void)
 847 {
 848         apic->read = xen_apic_read;
 849         apic->write = xen_apic_write;
 850         apic->icr_read = xen_apic_icr_read;
 851         apic->icr_write = xen_apic_icr_write;
 852         apic->wait_icr_idle = xen_apic_wait_icr_idle;
 853         apic->safe_wait_icr_idle = xen_safe_apic_wait_icr_idle;
 854
 855 #ifdef CONFIG_SMP
 856         apic->send_IPI_allbutself = xen_send_IPI_allbutself;
 857         apic->send_IPI_mask_allbutself = xen_send_IPI_mask_allbutself;
 858         apic->send_IPI_mask = xen_send_IPI_mask;
 859         apic->send_IPI_all = xen_send_IPI_all;
 860         apic->send_IPI_self = xen_send_IPI_self;
 861 #endif
 862 }
 863
 864 #endif
 865
 866 static void xen_clts(void)
 867 {
 868         struct multicall_space mcs;
 869
 870         mcs = xen_mc_entry(0);
 871
 872         MULTI_fpu_taskswitch(mcs.mc, 0);
 873
 874         xen_mc_issue(PARAVIRT_LAZY_CPU);
 875 }
 876
 877 static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
 878
 879 static unsigned long xen_read_cr0(void)
 880 {
 881         unsigned long cr0 = this_cpu_read(xen_cr0_value);
 882
 883         if (unlikely(cr0 == 0)) {
 884                 cr0 = native_read_cr0();
 885                 this_cpu_write(xen_cr0_value, cr0);
 886         }
 887
 888         return cr0;
 889 }
 890
 891 static void xen_write_cr0(unsigned long cr0)
 892 {
 893         struct multicall_space mcs;
 894
 895         this_cpu_write(xen_cr0_value, cr0);
 896
 897         /* Only pay attention to cr0.TS; everything else is
 898            ignored. */
 899         mcs = xen_mc_entry(0);
 900
 901         MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
 902
 903         xen_mc_issue(PARAVIRT_LAZY_CPU);
 904 }
 905
 906 static void xen_write_cr4(unsigned long cr4)
 907 {
 908         cr4 &= ~X86_CR4_PGE;
 909         cr4 &= ~X86_CR4_PSE;
 910
 911         native_write_cr4(cr4);
 912 }
 913
 914 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
 915 {
 916         int ret;
 917
 918         ret = 0;
 919
 920         switch (msr) {
 921 #ifdef CONFIG_X86_64
 922                 unsigned which;
 923                 u64 base;
 924
 925         case MSR_FS_BASE:               which = SEGBASE_FS; goto set;
 926         case MSR_KERNEL_GS_BASE:        which = SEGBASE_GS_USER; goto set;
 927         case MSR_GS_BASE:               which = SEGBASE_GS_KERNEL; goto set;
 928
 929         set:
 930                 base = ((u64)high << 32) | low;
 931                 if (HYPERVISOR_set_segment_base(which, base) != 0)
 932                         ret = -EIO;
 933                 break;
 934 #endif
 935
 936         case MSR_STAR:
 937         case MSR_CSTAR:
 938         case MSR_LSTAR:
 939         case MSR_SYSCALL_MASK:
 940         case MSR_IA32_SYSENTER_CS:
 941         case MSR_IA32_SYSENTER_ESP:
 942         case MSR_IA32_SYSENTER_EIP:
 943                 /* Fast syscall setup is all done in hypercalls, so
 944                    these are all ignored.  Stub them out here to stop
 945                    Xen console noise. */
 946                 break;
 947
 948         case MSR_IA32_CR_PAT:
 949                 if (smp_processor_id() == 0)
 950                         xen_set_pat(((u64)high << 32) | low);
 951                 break;
 952
 953         default:
 954                 ret = native_write_msr_safe(msr, low, high);
 955         }
 956
 957         return ret;
 958 }
 959
 960 void xen_setup_shared_info(void)
 961 {
 962         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 963                 set_fixmap(FIX_PARAVIRT_BOOTMAP,
 964                            xen_start_info->shared_info);
 965
 966                 HYPERVISOR_shared_info =
 967                         (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
 968         } else
 969                 HYPERVISOR_shared_info =
 970                         (struct shared_info *)__va(xen_start_info->shared_info);
 971
 972 #ifndef CONFIG_SMP
 973         /* In UP this is as good a place as any to set up shared info */
 974         xen_setup_vcpu_info_placement();
 975 #endif
 976
 977         xen_setup_mfn_list_list();
 978 }
 979
 980 /* This is called once we have the cpu_possible_mask */
 981 void xen_setup_vcpu_info_placement(void)
 982 {
 983         int cpu;
 984
 985         for_each_possible_cpu(cpu)
 986                 xen_vcpu_setup(cpu);
 987
 988         /* xen_vcpu_setup managed to place the vcpu_info within the
 989            percpu area for all cpus, so make use of it */
 990         if (have_vcpu_info_placement) {
 991                 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
 992                 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
 993                 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
 994                 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
 995                 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
 996         }
 997 }
 998
 999 static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1000                           unsigned long addr, unsigned len)
1001 {
1002         char *start, *end, *reloc;
1003         unsigned ret;
1004
1005         start = end = reloc = NULL;
1006
1007 #define SITE(op, x)                                                     \
1008         case PARAVIRT_PATCH(op.x):                                      \
1009         if (have_vcpu_info_placement) {                                 \
1010                 start = (char *)xen_##x##_direct;                       \
1011                 end = xen_##x##_direct_end;                             \
1012                 reloc = xen_##x##_direct_reloc;                         \
1013         }                                                               \
1014         goto patch_site
1015
1016         switch (type) {
1017                 SITE(pv_irq_ops, irq_enable);
1018                 SITE(pv_irq_ops, irq_disable);
1019                 SITE(pv_irq_ops, save_fl);
1020                 SITE(pv_irq_ops, restore_fl);
1021 #undef SITE
1022
1023         patch_site:
1024                 if (start == NULL || (end-start) > len)
1025                         goto default_patch;
1026
1027                 ret = paravirt_patch_insns(insnbuf, len, start, end);
1028
1029                 /* Note: because reloc is assigned from something that
1030                    appears to be an array, gcc assumes it's non-null,
1031                    but doesn't know its relationship with start and
1032                    end. */
1033                 if (reloc > start && reloc < end) {
1034                         int reloc_off = reloc - start;
1035                         long *relocp = (long *)(insnbuf + reloc_off);
1036                         long delta = start - (char *)addr;
1037
1038                         *relocp += delta;
1039                 }
1040                 break;
1041
1042         default_patch:
1043         default:
1044                 ret = paravirt_patch_default(type, clobbers, insnbuf,
1045                                              addr, len);
1046                 break;
1047         }
1048
1049         return ret;
1050 }
1051
1052 static const struct pv_info xen_info __initconst = {
1053         .paravirt_enabled = 1,
1054         .shared_kernel_pmd = 0,
1055
1056 #ifdef CONFIG_X86_64
1057         .extra_user_64bit_cs = FLAT_USER_CS64,
1058 #endif
1059
1060         .name = "Xen",
1061 };
1062
1063 static const struct pv_init_ops xen_init_ops __initconst = {
1064         .patch = xen_patch,
1065 };
1066
1067 static const struct pv_cpu_ops xen_cpu_ops __initconst = {
1068         .cpuid = xen_cpuid,
1069
1070         .set_debugreg = xen_set_debugreg,
1071         .get_debugreg = xen_get_debugreg,
1072
1073         .clts = xen_clts,
1074
1075         .read_cr0 = xen_read_cr0,
1076         .write_cr0 = xen_write_cr0,
1077
1078         .read_cr4 = native_read_cr4,
1079         .read_cr4_safe = native_read_cr4_safe,
1080         .write_cr4 = xen_write_cr4,
1081
1082         .wbinvd = native_wbinvd,
1083
1084         .read_msr = native_read_msr_safe,
1085         .write_msr = xen_write_msr_safe,
1086         .read_tsc = native_read_tsc,
1087         .read_pmc = native_read_pmc,
1088
1089         .iret = xen_iret,
1090         .irq_enable_sysexit = xen_sysexit,
1091 #ifdef CONFIG_X86_64
1092         .usergs_sysret32 = xen_sysret32,
1093         .usergs_sysret64 = xen_sysret64,
1094 #endif
1095
1096         .load_tr_desc = paravirt_nop,
1097         .set_ldt = xen_set_ldt,
1098         .load_gdt = xen_load_gdt,
1099         .load_idt = xen_load_idt,
1100         .load_tls = xen_load_tls,
1101 #ifdef CONFIG_X86_64
1102         .load_gs_index = xen_load_gs_index,
1103 #endif
1104
1105         .alloc_ldt = xen_alloc_ldt,
1106         .free_ldt = xen_free_ldt,
1107
1108         .store_gdt = native_store_gdt,
1109         .store_idt = native_store_idt,
1110         .store_tr = xen_store_tr,
1111
1112         .write_ldt_entry = xen_write_ldt_entry,
1113         .write_gdt_entry = xen_write_gdt_entry,
1114         .write_idt_entry = xen_write_idt_entry,
1115         .load_sp0 = xen_load_sp0,
1116
1117         .set_iopl_mask = xen_set_iopl_mask,
1118         .io_delay = xen_io_delay,
1119
1120         /* Xen takes care of %gs when switching to usermode for us */
1121         .swapgs = paravirt_nop,
1122
1123         .start_context_switch = paravirt_start_context_switch,
1124         .end_context_switch = xen_end_context_switch,
1125 };
1126
1127 static const struct pv_apic_ops xen_apic_ops __initconst = {
1128 #ifdef CONFIG_X86_LOCAL_APIC
1129         .startup_ipi_hook = paravirt_nop,
1130 #endif
1131 };
1132
1133 static void xen_reboot(int reason)
1134 {
1135         struct sched_shutdown r = { .reason = reason };
1136
1137         if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
1138                 BUG();
1139 }
1140
1141 static void xen_restart(char *msg)
1142 {
1143         xen_reboot(SHUTDOWN_reboot);
1144 }
1145
1146 static void xen_emergency_restart(void)
1147 {
1148         xen_reboot(SHUTDOWN_reboot);
1149 }
1150
1151 static void xen_machine_halt(void)
1152 {
1153         xen_reboot(SHUTDOWN_poweroff);
1154 }
1155
1156 static void xen_machine_power_off(void)
1157 {
1158         if (pm_power_off)
1159                 pm_power_off();
1160         xen_reboot(SHUTDOWN_poweroff);
1161 }
1162
1163 static void xen_crash_shutdown(struct pt_regs *regs)
1164 {
1165         xen_reboot(SHUTDOWN_crash);
1166 }
1167
1168 static int
1169 xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1170 {
1171         xen_reboot(SHUTDOWN_crash);
1172         return NOTIFY_DONE;
1173 }
1174
1175 static struct notifier_block xen_panic_block = {
1176         .notifier_call= xen_panic_event,
1177 };
1178
1179 int xen_panic_handler_init(void)
1180 {
1181         atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1182         return 0;
1183 }
1184
1185 static const struct machine_ops xen_machine_ops __initconst = {
1186         .restart = xen_restart,
1187         .halt = xen_machine_halt,
1188         .power_off = xen_machine_power_off,
1189         .shutdown = xen_machine_halt,
1190         .crash_shutdown = xen_crash_shutdown,
1191         .emergency_restart = xen_emergency_restart,
1192 };
1193
1194 /*
1195  * Set up the GDT and segment registers for -fstack-protector.  Until
1196  * we do this, we have to be careful not to call any stack-protected
1197  * function, which is most of the kernel.
1198  */
1199 static void __init xen_setup_stackprotector(void)
1200 {
1201         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1202         pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1203
1204         setup_stack_canary_segment(0);
1205         switch_to_new_gdt(0);
1206
1207         pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1208         pv_cpu_ops.load_gdt = xen_load_gdt;
1209 }
1210
1211 /* First C function to be called on Xen boot */
1212 asmlinkage void __init xen_start_kernel(void)
1213 {
1214         struct physdev_set_iopl set_iopl;
1215         int rc;
1216         pgd_t *pgd;
1217
1218         if (!xen_start_info)
1219                 return;
1220
1221         xen_domain_type = XEN_PV_DOMAIN;
1222
1223         xen_setup_machphys_mapping();
1224
1225         /* Install Xen paravirt ops */
1226         pv_info = xen_info;
1227         pv_init_ops = xen_init_ops;
1228         pv_cpu_ops = xen_cpu_ops;
1229         pv_apic_ops = xen_apic_ops;
1230
1231         x86_init.resources.memory_setup = xen_memory_setup;
1232         x86_init.oem.arch_setup = xen_arch_setup;
1233         x86_init.oem.banner = xen_banner;
1234
1235         xen_init_time_ops();
1236
1237         /*
1238          * Set up some pagetable state before starting to set any ptes.
1239          */
1240
1241         xen_init_mmu_ops();
1242
1243         /* Prevent unwanted bits from being set in PTEs. */
1244         __supported_pte_mask &= ~_PAGE_GLOBAL;
1245 #if 0
1246         if (!xen_initial_domain())
1247 #endif
1248                 __supported_pte_mask &= ~(_PAGE_PWT | _PAGE_PCD);
1249
1250         __supported_pte_mask |= _PAGE_IOMAP;
1251
1252         /*
1253          * Prevent page tables from being allocated in highmem, even
1254          * if CONFIG_HIGHPTE is enabled.
1255          */
1256         __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1257
1258         /* Work out if we support NX */
1259         x86_configure_nx();
1260
1261         xen_setup_features();
1262
1263         /* Get mfn list */
1264         if (!xen_feature(XENFEAT_auto_translated_physmap))
1265                 xen_build_dynamic_phys_to_machine();
1266
1267         /*
1268          * Set up kernel GDT and segment registers, mainly so that
1269          * -fstack-protector code can be executed.
1270          */
1271         xen_setup_stackprotector();
1272
1273         xen_init_irq_ops();
1274         xen_init_cpuid_mask();
1275
1276 #ifdef CONFIG_X86_LOCAL_APIC
1277         /*
1278          * set up the basic apic ops.
1279          */
1280         set_xen_basic_apic_ops();
1281 #endif
1282
1283         if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1284                 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1285                 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1286         }
1287
1288         machine_ops = xen_machine_ops;
1289
1290         /*
1291          * The only reliable way to retain the initial address of the
1292          * percpu gdt_page is to remember it here, so we can go and
1293          * mark it RW later, when the initial percpu area is freed.
1294          */
1295         xen_initial_gdt = &per_cpu(gdt_page, 0);
1296
1297         xen_smp_init();
1298
1299 #ifdef CONFIG_ACPI_NUMA
1300         /*
1301          * The pages we from Xen are not related to machine pages, so
1302          * any NUMA information the kernel tries to get from ACPI will
1303          * be meaningless.  Prevent it from trying.
1304          */
1305         acpi_numa = -1;
1306 #endif
1307
1308         pgd = (pgd_t *)xen_start_info->pt_base;
1309
1310         /* Don't do the full vcpu_info placement stuff until we have a
1311            possible map and a non-dummy shared_info. */
1312         per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1313
1314         local_irq_disable();
1315         early_boot_irqs_disabled = true;
1316
1317         xen_raw_console_write("mapping kernel into physical memory\n");
1318         pgd = xen_setup_kernel_pagetable(pgd, xen_start_info->nr_pages);
1319
1320         /* Allocate and initialize top and mid mfn levels for p2m structure */
1321         xen_build_mfn_list_list();
1322
1323         /* keep using Xen gdt for now; no urgent need to change it */
1324
1325 #ifdef CONFIG_X86_32
1326         pv_info.kernel_rpl = 1;
1327         if (xen_feature(XENFEAT_supervisor_mode_kernel))
1328                 pv_info.kernel_rpl = 0;
1329 #else
1330         pv_info.kernel_rpl = 0;
1331 #endif
1332         /* set the limit of our address space */
1333         xen_reserve_top();
1334
1335         /* We used to do this in xen_arch_setup, but that is too late on AMD
1336          * were early_cpu_init (run before ->arch_setup()) calls early_amd_init
1337          * which pokes 0xcf8 port.
1338          */
1339         set_iopl.iopl = 1;
1340         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1341         if (rc != 0)
1342                 xen_raw_printk("physdev_op failed %d\n", rc);
1343
1344 #ifdef CONFIG_X86_32
1345         /* set up basic CPUID stuff */
1346         cpu_detect(&new_cpu_data);
1347         new_cpu_data.hard_math = 1;
1348         new_cpu_data.wp_works_ok = 1;
1349         new_cpu_data.x86_capability[0] = cpuid_edx(1);
1350 #endif
1351
1352         /* Poke various useful things into boot_params */
1353         boot_params.hdr.type_of_loader = (9 << 4) | 0;
1354         boot_params.hdr.ramdisk_image = xen_start_info->mod_start
1355                 ? __pa(xen_start_info->mod_start) : 0;
1356         boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
1357         boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
1358
1359         if (!xen_initial_domain()) {
1360                 add_preferred_console("xenboot", 0, NULL);
1361                 add_preferred_console("tty", 0, NULL);
1362                 add_preferred_console("hvc", 0, NULL);
1363                 if (pci_xen)
1364                         x86_init.pci.arch_init = pci_xen_init;
1365         } else {
1366                 const struct dom0_vga_console_info *info =
1367                         (void *)((char *)xen_start_info +
1368                                  xen_start_info->console.dom0.info_off);
1369
1370                 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1371                 xen_start_info->console.domU.mfn = 0;
1372                 xen_start_info->console.domU.evtchn = 0;
1373
1374                 /* Make sure ACS will be enabled */
1375                 pci_request_acs();
1376
1377                 xen_acpi_sleep_register();
1378         }
1379
1380
1381         xen_raw_console_write("about to get started...\n");
1382
1383         xen_setup_runstate_info(0);
1384
1385         /* Start the world */
1386 #ifdef CONFIG_X86_32
1387         i386_start_kernel();
1388 #else
1389         x86_64_start_reservations((char *)__pa_symbol(&boot_params));
1390 #endif
1391 }
1392
1393 static int init_hvm_pv_info(int *major, int *minor)
1394 {
1395         uint32_t eax, ebx, ecx, edx, pages, msr, base;
1396         u64 pfn;
1397
1398         base = xen_cpuid_base();
1399         cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1400
1401         *major = eax >> 16;
1402         *minor = eax & 0xffff;
1403         printk(KERN_INFO "Xen version %d.%d.\n", *major, *minor);
1404
1405         cpuid(base + 2, &pages, &msr, &ecx, &edx);
1406
1407         pfn = __pa(hypercall_page);
1408         wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1409
1410         xen_setup_features();
1411
1412         pv_info.name = "Xen HVM";
1413
1414         xen_domain_type = XEN_HVM_DOMAIN;
1415
1416         return 0;
1417 }
1418
1419 void __ref xen_hvm_init_shared_info(void)
1420 {
1421         int cpu;
1422         struct xen_add_to_physmap xatp;
1423         static struct shared_info *shared_info_page = 0;
1424
1425         if (!shared_info_page)
1426                 shared_info_page = (struct shared_info *)
1427                         extend_brk(PAGE_SIZE, PAGE_SIZE);
1428         xatp.domid = DOMID_SELF;
1429         xatp.idx = 0;
1430         xatp.space = XENMAPSPACE_shared_info;
1431         xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
1432         if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1433                 BUG();
1434
1435         HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
1436
1437         /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1438          * page, we use it in the event channel upcall and in some pvclock
1439          * related functions. We don't need the vcpu_info placement
1440          * optimizations because we don't use any pv_mmu or pv_irq op on
1441          * HVM.
1442          * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1443          * online but xen_hvm_init_shared_info is run at resume time too and
1444          * in that case multiple vcpus might be online. */
1445         for_each_online_cpu(cpu) {
1446                 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
1447         }
1448 }
1449
1450 #ifdef CONFIG_XEN_PVHVM
1451 static int __cpuinit xen_hvm_cpu_notify(struct notifier_block *self,
1452                                     unsigned long action, void *hcpu)
1453 {
1454         int cpu = (long)hcpu;
1455         switch (action) {
1456         case CPU_UP_PREPARE:
1457                 xen_vcpu_setup(cpu);
1458                 if (xen_have_vector_callback)
1459                         xen_init_lock_cpu(cpu);
1460                 break;
1461         default:
1462                 break;
1463         }
1464         return NOTIFY_OK;
1465 }
1466
1467 static struct notifier_block xen_hvm_cpu_notifier __cpuinitdata = {
1468         .notifier_call  = xen_hvm_cpu_notify,
1469 };
1470
1471 static void __init xen_hvm_guest_init(void)
1472 {
1473         int r;
1474         int major, minor;
1475
1476         r = init_hvm_pv_info(&major, &minor);
1477         if (r < 0)
1478                 return;
1479
1480         xen_hvm_init_shared_info();
1481
1482         if (xen_feature(XENFEAT_hvm_callback_vector))
1483                 xen_have_vector_callback = 1;
1484         xen_hvm_smp_init();
1485         register_cpu_notifier(&xen_hvm_cpu_notifier);
1486         xen_unplug_emulated_devices();
1487         x86_init.irqs.intr_init = xen_init_IRQ;
1488         xen_hvm_init_time_ops();
1489         xen_hvm_init_mmu_ops();
1490 }
1491
1492 static bool __init xen_hvm_platform(void)
1493 {
1494         if (xen_pv_domain())
1495                 return false;
1496
1497         if (!xen_cpuid_base())
1498                 return false;
1499
1500         return true;
1501 }
1502
1503 bool xen_hvm_need_lapic(void)
1504 {
1505         if (xen_pv_domain())
1506                 return false;
1507         if (!xen_hvm_domain())
1508                 return false;
1509         if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1510                 return false;
1511         return true;
1512 }
1513 EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1514
1515 const struct hypervisor_x86 x86_hyper_xen_hvm __refconst = {
1516         .name                   = "Xen HVM",
1517         .detect                 = xen_hvm_platform,
1518         .init_platform          = xen_hvm_guest_init,
1519 };
1520 EXPORT_SYMBOL(x86_hyper_xen_hvm);
1521 #endif